Corresponding author: L. M. Beal, Rosenstiel School of Marine and Atmospheric Science, University of Miami, 4600 Rickenbacker Causeway, Miami, FL 33149, USA. (email@example.com)
 In situ measurements are used, together with sea surface height data, to study the development and variability of the Great Whirl (GW), a large quasi-stationary anticyclone that appears off the coast of Somalia during the southwest monsoon season. We find that anticyclonic circulation indicative of the GW appears on average in April, almost two months before the onset of the southwest monsoon winds. This early initiation is coincident with the arrival of annual Rossby waves at the western boundary. With the onset of the monsoon winds in early June, the GW-proper intensifies quickly, remaining at its peak throughout July, August, and September, and dissipating about one month after the winds have died. The GW is present for an average 166 ± 30 days per year and the position of its northern flank, close to 9°N, coincides with the latitude of zero wind stress curl. The intraseasonal variability of the GW is intense as a result of mutual advection with one to three flanking cyclones, which accompany the GW 70% of the time and tend to circulate clockwise around it. There is no consistent seasonal pattern for the development or dissipation of the GW, although movement to the southwest is common toward the end of the season. The GW of 1995 deepened from 200 m in June to over 1000 m in September, and strengthened from a swirl transport of 10 to 60 Sv. Cool waters in its core resulted from advection via the Somali Current and some local vertical mixing.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 In his Sailing Directory for the Indian Ocean of 1876, Alexander Findlay described the southwest monsoon of the Indian Ocean as “…an exceptional phenomenon, in direct opposition to the normal aerial system, and which, by the forces which call it into existence, is made one of the most formidable manifestations of the power of the wind that can be adduced in any part of the globe.” The monsoon winds that Findlay  described can reach 14 m s–1 and annually reverse the entire Arabian Sea circulation, forcing strong, southerly Ekman velocities in the ocean interior and a northward western boundary current along the Somali coast. In addition, a large, quasi-stationary, anticyclonic eddy spins up off the coast of Somalia around latitude 8°N. This feature was first described by Lieutenant A. D. Taylor of the Royal (British) Navy as a “great whirl of current” [Findlay, 1876], and has subsequently become known as simply The Great Whirl. Mariners have long been wary of the phenomenon, which exhibits swirling surface currents up to 2.5 m s–1, as intense as any current in the world's oceans.
 Owing to the seasonality of the forcing, the Somali Current and Great Whirl (GW) do not have a representative time-mean signature; instead their circulations evolve over the course of the southwest monsoon. Schott  describes their development as a series of stationary flow patterns linked by short periods of rapid change, and more recently Schott and McCreary [2001, Figure 32] imply a more steadily evolving circulation. Briefly, the historical understanding is as follows. In early May a weak and shallow, northward coastal current begins to flow across the equator, probably forced by southern hemisphere easterlies [McCreary and Kundu, 1988]. As alongshore winds build in late May, the current reaches about 3°N, where it separates from the coast and wraps back across the equator, forming the shallow Southern Gyre [Düing et al., 1980; Jensen, 1991]. A wedge-shaped region of upwelling forms along the northern flank of this Gyre. At this time there is also a continuation of the boundary current, the Somali Current, farther north, driven by the local alongshore winds. In June the monsoon winds begin in earnest, with strong southwesterlies (up to 14 m s–1) penetrating offshore at about 9°N (the Findlater Jet [Findlater, 1969]), and a large anticyclone, the GW, spins up between 5°N and 10°N [Leetmaa et al., 1982]. Another cool wedge forms along its northern arm and subsequently a component of the Somali Current continues northward, past the recirculating GW, to flow through the Socotra Passage [Fischer et al., 1996]. The current can penetrate beyond the Horn of Africa and across the mouth of the Gulf of Aden [Beal and Chereskin, 2003], although it is not known how frequently it may do so. During August and September the Somali Current and the GW strengthen [Fischer et al., 1996], while the Southern Gyre weakens. There is some observational and model evidence [Evans and Brown, 1981; Swallow et al., 1983; Luther and O'Brien, 1989] that the Southern Gyre migrates northward in September (before the monsoon winds weaken) to coalesce with the GW, but it is not clear how typical this behavior is. On occasion the Somali Current may be found offshore of the GW [Fischer et al., 1996] and a third anticyclone, the Socotra Eddy, is sometimes found to the east of Socotra [Bruce, 1983]. At the end of the monsoon in late September the Somali Current and GW collapse.
 Studying the variability of the GW is difficult with direct measurements, because resolving the fast pace of its development requires many repeat observations. However, some modeling studies have given insights into what the important processes might be that govern the formation and decay of the GW. In general, the existence of the Southern Gyre and GW, rather than a continuous boundary jet along the Somali coast, is attributed to the 45° slant of the coastline [Cox, 1979; McCreary and Kundu, 1988], which arrests the northward propagation of eddies and the subsequent transition to a more linear boundary flow. McCreary and Kundu  found that a GW appears as a result of local alongshore winds and that the larger-scale curl associated with the offshore weakening of the Findlater Jet is an unnecessary condition. Jensen  investigated the formation and decay of the GW in detail, and suggests that its formation is due to barotropic instabilities brought about by a sign change in the horizontal potential vorticity gradient. In other words, his 3.5 layer model shows that the Somali Current advects a potential vorticity anomaly to the north (south equatorial waters), until it creates a relative vorticity gradient large enough to satisfy conditions for instability. At this point the GW is formed and slowly deepens by downward momentum transfer so that anticyclonic motion is seen at 850 m depth after 18 days.
 The interannual variability of the position and strength of the GW is not well understood. Luther and O'Brien  found its variability to be almost entirely independent of the internal ocean physics and therefore solely determined by the variability of the wind forcing. In contrast, Wirth et al.  found that most of its variability is caused by the chaotic nature of the ocean dynamics. The eventual decay of the GW is described plausibly by Jensen  as being due to baroclinic instability brought on by its interaction with an adjacent cyclone, which is generated by shear between the GW and the island of Socotra. The two eddies mutually advect southeastward, which tilts the axis of the Whirl, thereby increasing vertical shear and causing the flow to become unstable.
 A lack of observations in the Arabian Sea, particularly over the past 10 or 15 years owing to Somali piracy, combined with the seasonality of the monsoon forcing, means that there are insufficient in situ data to resolve the annual evolution of the GW and its variability. In this paper we use in situ data from the well-observed 1995 southwest monsoon, combined with almost two decades of sea surface height data from satellite altimeter, to study the seasonal evolution of the GW, both in 1995 and climatologically, as well as its intraseasonal to interannual variability.
 The GW was traversed twice during the summer monsoon of 1995, as part of the World Ocean Circulation Experiment Hydrographic Program (WHP). The first occupation was at the beginning of June (NOAA WHP repeat line I1, 5/6/95, stations starting 130, Figure 1) and the second in mid-September (WHP line I1, 12 September 95, stations starting 900, Figure 1). During the first section in June, stations were occupied zonally along 8.5°N. In September the stations run perpendicular to the topography of the continental slopes and the marine ridges, with zonal sections along 8.5°N across the Somali and Arabian Basins. During both occupations full-depth measurements of salinity, temperature, pressure, oxygen, and velocity were taken on station using a combined Conductivity Temperature Depth/Lowered Acoustic Doppler Current Profiler (CTD/LADCP) package. Continuous velocity profiles down to almost 400 m depth were collected while underway, using a hull-mounted, Acoustic Doppler Current Profiler (SADCP).
 The SADCP was an RD Instruments 150 kHz, narrowband instrument. Navigation was by military P(Y)-code GPS and as a result the uncertainty in the 10 min averaged absolute current due to errors in ship positioning is 1.2 cm s–1. Heading was collected via gyrocompass with corrections from an Ashtech 3DF GPS attitude sensor, for a heading accuracy of about 0.1°. Overall the estimated errors in 10 min averaged absolute currents for September are less than 2 cm s–1 [Chereskin et al., 1997]. The LADCP was a 150 kHz, broadband instrument, mounted on the CTD rosette. We used the UH (Firing) processing technique to build a full depth profile from the LADCP data [Firing and Gordon, 1990; Fischer and Visbeck, 1993]. For a cast of 2500 m depth, the accumulated error is about 2.4 cm s–1 [Beal and Bryden, 1999]. Navigational inaccuracies and the data-gap, which occurs at the bottom of each cast make the combined estimated error for absolute current from LADCP order 5 cm s–1 (dependent on depth). During the September cruise there was LADCP failure below 1200 m on three stations in the GW (906–908). The loss of deep data on these stations degrades the absolute velocity to uncertainties of order 10 cm s–1. Salinity and temperature were sampled, processed, and calibrated to WOCE standards on the June cruise (WOCE CTD specs for temperature: accuracy 0.002°C and precision 0.0005°C, and for salinity: accuracy 0.002 and precision 0.001). However, instrumental problems during the September line doubled salinity errors (temperature: accuracy 0.002°C and precision 0.001°C, and salinity: accuracy 0.004 and precision 0.002). Geostrophic velocities were calculated from the CTD data and referenced using the linearly-interpolated, depth-mean velocity from the two LADCP profiles either side of each geostrophic profile, following Beal and Bryden . In the case of the three short LADCP profiles, the geostrophic profiles were referenced to the direct velocity at 1200 m depth.
 The GW can be observed over the time between the two hydrographic occupations and on interannual time scales by using 18 years of satellite sea surface height data (1993 to 2010). We have used 7 day mapped sea level anomalies generated by AVISO the CLS Space Oceanography Division, France. Mapping was achieved using an improved space/time objective analysis method [Le Traon et al., 1998], which takes into account long wavelength errors (i.e., correlated noise). The spatial and temporal correlation functions vary with latitude and for our region of interest are: a zero crossing (ZC) at 350 km zonally, 250 km meridionally, and a Gaussian with an e-folding time of 15 days, respectively. For each grid point, data inside a time window of ± 10 days for TOPEX/POSEIDON and ± 18 days for ERS, and a space window of ± 3 ZC are used. Features of less than 250 km in size are expected to be followed poorly (fade in and out) in the mapped sea level anomalies maps, because the raw TOPEX track data are spaced 2° apart. A mean or reference sea surface height (SSH) field was obtained from Maximenko and Niiler , who combined large-scale mean sea level from the GRACE mission with mesoscale sea level tilt derived from global drifter, satellite altimeter, and wind data.
3 Direct Observations
 According to the Indian Monsoon Index, as defined by Wang and Fan , the 1995 southwest monsoon was neither unusually strong nor weak. It was characterized by a double onset, in which a small southwesterly wind event from beginning to mid-May was followed by a two week lull in the winds [Flatau et al., 2001]. The monsoon proper began on 1 June, with persistent southwesterly winds (11 m s–1 at 237° on average) blowing until 15 September. Thus, our first set of measurements taken in the GW were one week after the onset of the monsoon winds and the second set were right at the end of the monsoon, but before the winds had weakened significantly. There was no direct evidence of a Somali Current in June, owing to a lack of observations close to the coast (permission to work in Somali waters was not obtained). In September the Somali Current was deep-reaching with a substantial transport of 37 ± 5 Sv [Beal et al., 2003].
3.1 Size and Intensity
 In general, the GW sits at the apex of the Somali Basin (Figure 1), which is bounded in the west by the continental slope of Somalia and to the east is separated from the Arabian Basin by the Carlsberg Ridge, which peaks to about 2500 m depth. To the north is the island of Socotra, which is joined to the Horn of Africa by a shallow shelf area no greater than 200 m deep, cut by a single narrow passage aligned southeast-to-northwest with a sill depth of 1000 m, at about 51.6°E, 12.2°N.
 In June the GW is 350 km wide with maximum surface velocities of 120 cm s–1 and penetrates no more than 200 m deep (Figures 1 and 2). Adjacent is a weaker anticyclone farther offshore. Density (neutral) surfaces slope upward toward Somalia from the center of the GW, so that there is no separation between it and the Somali Current in this section [Beal and Chereskin, 2003]. Cooler waters from below the top of the pycnocline (TTP) outcrop at the coast. The TTP deepens to 100 m in the center of the whirl [Chereskin et al., 2002]. The velocity structure tends to follow the shape of the density surfaces, particularly toward the western boundary, so that the speed of the flow attenuates quickly (high vertical shear) at the TTP. Comparing direct and LADCP-referenced geostrophic profiles of velocity (see Beal and Bryden  for a full description of the method), the circulation is found to be close to geostrophic balance. The swirl transport is 10 Sv, with a small northward net transport of 2 Sv.
 By September the GW has grown considerably and measures 540 km across, occupying almost the entire width of the Somali Basin at this latitude (Figures 1 and 2). It has a large ageostrophic component, as evidenced from differences between geostrophic and direct velocities of up to 70 cm s–1 at the surface (not shown) and a calculated Rossby number over 0.1. The depth of the TTP has almost tripled to 280 m in the center of the GW, while the underlying pycnocline is considerably thinner [Chereskin et al., 2002]. The broad, deep bowl of neutral density surfaces [Jackett and McDougall, 1997] depicting the GW has shoulders on either side, indicating flanking cyclonic flows (green contours in Figure 2 and 4), and a separation from the Somali Current at the western boundary [Beal and Chereskin, 2003]. Between the GW and the Somali Current there is westward flow (Figure 2 and 3), clearest below 800 m. As will be seen later in the satellite altimeter data, flanking cyclones are a common feature of the GW.
 In terms of penetration of the GW, there are velocities of significant magnitude (larger than expected errors) in waters up to 3000 m deep. Figure 3 shows how anticyclonic rotation is coherent throughout the water column, with little shifting in the vector directions (although the northward arm of the GW is missing below 1200 m due to instrument failure). However, given the stratification of the water column, it seems unlikely that these currents could result from a deep penetration of the wind-forced GW. Furthermore, water mass signatures indicate that trapping is limited to the upper 600 m. Current meter data from the region show fluctuating deep currents of 5–10 cm s–1 at 3000 m depth [Dengler et al., 2002], with a period of about 40 days. This is the same period found in local SSH by Brandt et al., 2003, who suggested these fluctuations are free Rossby waves emanating from internal instabilities of the GW. Hence, given the right timing, these Rossby waves could account for our deep observations. Mean currents over the period July to September 1995 from a nearby moored array [Schott and McCreary, 2001] indicate that the GW reached down to 1000 m on average, with speeds of 10 cm s–1.
 For a swirl transport in September we take the anticyclonic circulation offshore of the Somali Current, which fills the Somali Basin from station 904 to station 913 (Figure 2), down to the depth of the γ = 27.75 neutral density surface (about 1500 m), which marks the transition from intermediate to deep waters (Figure 4). This avoids large errors from extrapolated profiles on those stations with missing deep data (906 to 908). The swirl transport is 60 Sv, with a net northward flow of 30 Sv.
3.2 Water Masses
 The water mass signatures of the GW are of interest in two regards. First in terms of its connectivity to the Somali Current, and second because of its unusual characteristic as an anticyclone with a cold core. In June, and more distinctly in September, surface and subsurface waters in the Great Whirl and Somali Current are cooler and fresher than those immediately to the east (Figure 1), as found previously during the southwest monsoon [Schott et al., 2002]. This phenomenon is illustrated clearly by looking at the cross-sections of salinity presented in Figure 4, which shows that the salty, subsurface layer of the Arabian Sea surface water found along 8.5°N is not found within the GW in June (west of station 139). In September, the water properties of the GW and Somali Current are also much fresher than in the interior.
 A T-S diagram (Figure 5) more clearly shows the difference between waters in the GW and in the interior. The interior waters here are defined as those from the Carlsberg Ridge across to the Chagos-Laccadives Ridge at 70°N (I1R: stations 138–161 and I1: stations 914–942), spanning the Arabian Basin while omitting the eastern boundary at India, where the fresh signature of Bay of Bengal water enters the basin seasonally and muddles the picture. It is evident then, that both the GW and Somali Current contain considerably fresher waters than those in the interior and we deduce (as did Schott ) that these waters are advected from a southerly source. In addition, between June and September there is water mass modification over the top 200 m or more, an indication of vertical mixing. There is some evidence of further changes down to about σ0 = 27.0 at around 400 m. A knee of Red Sea water is apparent in both curves at 10°C, indicating that, on average, waters flow from the north at these depths (presumably through the deep passage between Socotra and the horn of Africa) throughout the season. This suggests that the trapping depth of the GW, i.e., that depth to which anomalous water masses are trapped within the feature, is somewhere between 400 and 600 m depth.
 In conclusion, we observe a dramatic intensifying and deepening of the GW from June to September. While the feature is constrained above the TTP in June, by September a coherent, anticyclonic circulation is observed to reach down to at least 1200 m depth. Vertical mixing appears to help maintain the cool core waters of the GW, which must be fed by cool Somali Current waters from the south for a considerable proportion of the season.
4 Variability of the Great Whirl from Sea Surface Height Data
 Animations of sea surface height data give insight into the spatial and temporal variability of the Great Whirl over a wide range of time scales. For the 1995 southwest monsoon, we can look at the variability of the GW between the in situ occupations described above. Using the 18 year data set (1993–2010) we can create monthly composites to examine the average seasonal development of the GW, as well as quantify its interannual variability.
4.1 Ground-truthing the SSH Data
 The ability of the SSH fields to resolve the variability of the GW is dependent on the spatial and temporal resolution of the data. In particular, mesoscale features when mapped may appear to fade in and out, when in fact they are simply advecting between TOPEX lines. Hence, the degree of variability may be falsely enhanced and real oceanic features may be poorly represented. To test how well the SSHs describe the GW and the eddy field surrounding it, we matched up five images from the 1995 southwest monsoon season with in situ data taken during WOCE. In addition to the sections at 8°N described previously, the Meteor crossed the GW twice, and the Malcolm Baldridge once (again). Figure 6 shows shipboard ADCP data in the region of the GW from each of these cruises, superposed on SSH anomaly (SSHA).
 The five panels of Figure 6 are placed in chronological order from top left to bottom right. The first crossing of the GW in 1995 is the early June crossing that we show in Figure 2. The ADCP velocities follow SSHA contours closely at the northern tip of the developing GW and also reflect other features, such as the anticyclonic flow north of Socotra. At the end of June, the Meteor cruise track criss-crosses the region, traversing the Socotra eddy and the eastern flank of the GW. By August (bottom panel) the GW and Socotra eddy are considerably stronger and the second Baldridge cruise crosses the Socotra eddy for a third time. At the beginning of September the second Meteor cruise captures many sections through the GW. Velocity vectors are parallel to SSHA contours almost everywhere and the mid-point of the GW is colocated in the two types of measurements. The strong flanking cyclones (as seen in the section data) and the offshore northward jet are also illustrated by SSHA and velocities alike. Finally, during the Knorr cruise in mid-September there also appears to be good agreement between satellite and in situ features.
 A more quantitative comparison of the SSH fields with direct velocity measurements can be made by using the gradients, SSHxy to estimate surface currents. Figure 7 presents the v-component from SSH, Vgeo = g/f SSHx, together with in situ northward ADCP velocities, VADCP, for the five cruises of 1995 (shown in Figure 6). Owing to the many changes in direction and orientation of the cruise tracks, velocities are shown simply as a function of ADCP profile-index and each profile represents a 15 min average. There are gaps in Vgeo where there is no SSH signal close to the coast and the island of Socotra (although it is interpolated in the maps of Figure 6). The spatial and temporal scales of the mesoscale features match well, suggesting that the rapid variability of features seen in SSH imagery is real and not the effect of mapping over data gaps. The relatively large time and space scales of the SSH information do act to smooth over smaller-scale features, as can be seen most clearly in the second panel of Figure 7. Table 1 gives a summary of the statistics of ADCP (VADCP) versus SSH-derived (Vgeo) velocities. Vgeo consistently underestimates in situ values, which is no surprise since in addition to smoothing, Ekman, near-inertial, and nonlinear terms are unaccounted for. A linear fit of Vgeo and VADCP indicates an underestimation of 60% to 70% on average. The mean bias of about 8 cm s–1 in the V component (Table 1) probably reflects the southward Ekman transport over the region [Chereskin et al., 2002]. We attempted nonlinear corrections to Vgeo, both cyclostrophic and advective terms, but with mixed results. The overall fit to the ADCP was only improved marginally and not as much as anticipated. In summary, SSH images show real ocean features, although significantly reduced in amplitude, and can be confidently studied for insights into the variability of the GW and the regional mesoscale eddy field.
Table 1. Statistics of the Comparison Between In Situ and SSH-derived Velocities Within the Region of the Great Whirl During the 1995 Summer Monsoon
RMS differences (m/s)
4.2 Analysis of SSH Fields
 Eighteen years of SSH animations, from 1993 through 2010, were studied. The GW was defined as a high >35 cm in SSH between 5°N and 10°N and west of 55°E. Various criteria were noted for each image, such as when the GW appeared and dissipated, whether it was anisotropic in size, how many cyclones flanked it, and whether it splits or merges with another eddy. We also qualitatively examined the sequence of events leading up to its dissipation to see if flanking cyclones played a role, as suggested by Jensen . Identifying a signal as the Socotra eddy was more difficult and open to some subjectivity. In essence, the SSH data show that the GW is rapidly varying in position and shape, largely owing to mutual advection with adjacent cyclones and occasionally anticyclones. Our findings are in contrast to many previous studies, which have proposed a more slowly varying GW which may respond predominantly to wind forcing [Schott, 1983; Luther and O'Brien, 1989; Schott and McCreary, 2001]. Figure 8 illustrates the prevalence of the cyclones flanking the GW and their affect on its shape and position.
 The top panel of Figure 8 shows a zero lag correlation of SSH with a high at 8°N, 53°E, the mean position of the SSH maximum from June to September. The resulting pattern indicates that the GW is strongly associated with cyclonic circulation to its northeast and south. Investigations at different lags confirm that these cyclones are not propagated from the east (not shown). This finding supports the numerical study of Jensen , where cyclones are generated by shear between the GW and the island of Socotra. The lower panels of Figure 8 show 6 years of case studies from the SSH animations, taken during the month of August. Cyclones are typically advected clockwise around the GW, as seen most clearly in 1996 and 1998. Another common phenomenon is the merging of the GW with another anticyclone to the northeast, nominally the Socotra eddy. This merging occurs in four of the six case studies shown (years 1996, 1998, 1999, and 2000) and about 50% of the time over the full 18 year record.
 The short-term variability of the GW, as seen in the SSH case studies, can also affect the Somali Current. The two SSH/ADCP fields shown in the right-hand column of Figure 6 are 7 days apart and the GW changes its orientation from east-west to north-south over this short time. In the first image, the GW appears to block the Somali Current at the western boundary and deflect it into a strong offshore jet at about 55°E. This is corroborated by in situ data from Meteor, which Schott and McCreary  interpret as an evolution of the Somali Current system with the advancement of the monsoon. Yet, we see here that in the second image a week later, the offshore jet is gone. By synthesizing these 1995 data sets and examining the SSH animations, we come to the conclusion that summarizing the monsoon circulation as an organized pattern of changes that are similar in any given season, as in the review of Schott and McCreary , may be misleading. Instead, the GW varies rapidly and evolves rather stochastically, largely as a result of mutual eddy advection. Wirth et al.  also found highly stochastic variability in the GW in a 1/3° regional model.
 This kind of rapid variability is evident throughout all of the 18 seasons studied in the SSH animations, with the GW moving by up to 2° within 14 days. Between 1993 and 2010 we find that the GW is present on average 166 ± 30 days per year. It appears anytime between April and June (23 May ± 28 days) and typically remains until the beginning of November (3 November ± 23 days), although in one year it disappears in August, while in another it lasts until December. Its shape is approximately circular 33% of the time and anisotropic the remainder of the time, with no preference for the direction of its major axis. There are cyclones flanking the GW about 70% of the time, typically appearing in late June and remaining for the rest of the season. Between one and three cyclones are common. Most often they are formed between the GW and the island of Socotra to the north, as predicted by Jensen , and from there circulate anticyclonically around the GW. A second anticyclone, nominally the Socotra eddy, is present in the region for an average 75 ± 31 days per year. Bruce  originally named this anticyclone, noting its persistence to the east of the Island of Socotra, however it often spins up far to the south and east, advecting westward and only sometimes turning northward to stall by its island namesake for a period of time. The Socotra eddy exhibits greater variability in size, shape, and position than the GW and is often difficult to identify from other anticyclones in the region. In half the years studied, as seen in Figure 8, it interacts and merges with the GW. Sometimes it drifts west toward the end of the season after the GW has disappeared, to take its place off the coast of Somalia for a few weeks before eventually spinning down. Besides the GW, its flanking cyclones, and the Socotra eddy, other eddy features appear and disappear within the region and are difficult to track.
 In the past it has been suggested that the GW dissipates through a process of emergence with the Southern Gyre as the monsoon winds weaken [Swallow et al., 1983; Luther and O'Brien, 1989], or through growing baroclinic instability caused by interaction with a flanking cyclone [Jensen, 1991]. We observed that most years the GW moves southwest before it dissipates. For half the years studied an intensifying cyclone appears to push the GW southward, or split it from the east, toward the end of the season. However, statistically, interaction with cyclones does not lead to an earlier destruction of the GW. On average, the GW outlasts the winds by over a month, presumably due to inertia, dissipating on 3 November ± 23 days, compared to the mean monsoon end on 21 September [Wang and Fan, 1999].
 Looking for some seasonal consistency, we calculated monthly composites of absolute SSH, shown in Figure 9. These represent the climatological mean annual development of the GW. Corroborating our statistical study of the animations, the monthly composites show that the GW (again defined as a high in SSH between 5°N and 10°N and west of 55°E) appears as early as April, well before the climatological onset of the southwest monsoon winds on 1 June [Schott and McCreary, 2001]. Its appearance coincides with the arrival of a westward-propagating ridge of high SSH at the coast of Somalia (Figure 9). Brandt et al.  and McCreary et al.  have shown that this signal is associated with annual Rossby waves (first and second baroclinic mode), which radiate from the tip of India in November. Solving the dispersion relation for linear Rossby waves, using the characteristics of the incoming waves from Brandt et al. , we find that the length scale for short Rossby waves reflected at the boundary is between 60 and 70 km for a coastline slant of 45°, not inconsistent with the observed pattern of SSH. Moreover, in their 2.5 layer model, McCreary et al.  find that the arrival of annual Rossby waves causes northward flow at the boundary in April and May, against the prevailing winds. Therefore, the evidence points to an initiation of the GW at the boundary in April, as a result of remote forcing transmitted by annual Rossby waves. Further study is necessary to confirm this and to determine how much this early initiation affects the intensification of the GW proper once the southwest monsoon begins in June.
 The monthly composites (Figure 9) show that the GW does not intensify significantly past July and is strongest throughout the height of the southwest monsoon, in July, August, and September (Figure 9). There is some evidence for a Socotra eddy in the monthly composite fields, to the north and east of the GW, but it is clearly not stationary. The mean field of October suggests that at this point in time the Socotra eddy is stronger than the GW. Overall, the mean position of the northern flank of the GW is close to 9°N, coincident with the axis of the Findlater jet.
 Regarding interannual variability, Figure 10 shows a time series of GW intensity compared to the dominant climate modes over the Indian Ocean, the Indian Monsoon (IMI) [Wang and Fan, 1999] and Dipole Mode Indices (DMI) [Saji et al., 1999]. We define GW intensity as the mean absolute SSH over the same region used above, in each weekly SSH field. The IMI and DMI are taken as published, but normalized to the same scale as the SSH variability. Following the intensity of the GW (black line in Figure 10, top panel) through each season we see that there is no clear temporal separation between the onset of the precursor GW in April and the GW-proper in June. Although in many years there is a drop in intensity after the initial peak, in most summers there are multiple minima and in some cases, such as 2007, the greater minimum occurs at the height of the monsoon. In year 2000 there is no early onset and no minima during the summer. Hence, we could not find an objective way to separate these two phenomena, and indeed it appears that in most cases the initiation of the GW in April acts as a preconditioning for the monsoon circulation, although further study would be necessary to fully demonstrate this.
 We also define an annual, or summer intensity, based on the intensity of the GW averaged over July, August, and September, during its peak strength (Figure 10, bottom panel). Weekly values of GW intensity can be compared to the DMI and annual values of its summer intensity can be compared to IMI. There is no significant correlation between these time series, although the 1997/98 El Niño stands out as a period of high DMI and GW intensity, combined with the weakest southwest monsoon of the 20 year record. Also, we note that the merging of the GW with the Socotra eddy occurs in 1996, 1998, 1999, and 2000 (Figure 8), which are coincidentally La Niña years. There is significant coherence of the DMI and GW series at periods of 2 years, half year, and 40 days, probably related to common forcings at these periods (not shown). The latter period has been noted by Brandt et al., 2003, who suggested it emanates from instability processes of the GW itself, and also by Zhang  who related it to large-scale patterns of tropical Indian atmospheric variability, known as the Madden-Julian Oscillation.
 Eighteen years of satellite altimeter data, synthesized with in situ velocity measurements from the 1995 southwest monsoon season, show that the intraseasonal variability of the GW is strong and there is no fixed pattern of development or dissipation seasonally. Short-term variability of the position and shape of the GW is largely the result of mutual advection with one to three flanking cyclones. These cyclones are correlated with the GW, often spinning up between the GW and the island of Socotra, and tend to circulate clockwise around it. In addition to the associated cyclones, a second anticyclone, nominally the Socotra eddy, is often present to the northeast of the GW and in half the seasons studied, merges with it. Overall, the GW is present for 166 ± 30 days per year and the associated cyclones for 70% of this time. The position of its northern flank close to 9°N is notably the latitude of the zero wind stress curl along the axis of the Findlater jet.
 On the seasonal average the GW, or its precursor, appears as early as April, two months before the onset of the southwest monsoon winds. This initiation of anticyclonic circulation is coincident with an influx of vorticity from the arrival of annual first- and second-mode baroclinic Rossby waves at the Somali coast. Although further study is needed to confirm the link, remote forcing has been implicated before in a numerical study of the monsoon circulation [McCreary et al.,1993], where northward flow appeared along the Somali coast in April, against the prevailing winds. It is interesting to consider that this remote forcing pre-conditions the Arabian Sea before the arrival of the monsoon forcing and therefore may have an influence on the strength of the ensuing monsoon circulation. Once the southwest monsoon winds set in at the beginning of June the GW strengthens rapidly and remains at its peak throughout July, August, and September, typically dissipating in early November over a month after the winds have died.
 During the 1995 southwest monsoon in situ velocities from five independent cruises were found to be well correlated with geostrophic estimates from mapped SSH (r > 0.84), but were underestimated by up to 70%. A mean seasonal bias of 8 cm s–1 was attributed to southward Ekman currents. The GW was present for about 200 days with maximum measured surface velocities of 2 m s–1 in mid-September. Its vertical penetration was 200 m in June and over 1000 m in September, but with a trapping depth of about half this, as indicated by water mass signatures. Its swirl transport was estimated to be 10 Sv in June and over 60 Sv in September. Cool core waters of the GW were sourced from the Somali Current and may have been maintained by vertical mixing over the top 200 m of the water column, as evidenced by homogenized water mass properties. The GW of 1995 merged with the Socotra eddy in early October and eventually dissipated one month later.
 We are grateful for the involvement of Eric Firing, who provided the SSH-derived velocities and helped improve this manuscript. We acknowledge the officers and crew of the R/V Knorr and R/V Baldridge. The altimeter products were produced by Ssalto/Duacs and distributed by Aviso, with support from Cnes (http://www.aviso.oceanobs.com/duacs/). The initial parts of this study were done with support from NSF grants OCE99-06776 and OCE99-07458.